qwen-1.8b-adversarial-179k

This model is a fine-tuned version of Qwen/Qwen1.5-1.8B trained on 179,378 adversarial prompts.

Model Details

• Base Model: Qwen/Qwen1.5-1.8B
• Training Data: 179,378 adversarial examples
• Training Time: 2h 51m on NVIDIA A100
• Final Loss: 1.973
• Model Size: 3.42 GB

Training Configuration

• Batch Size: 32
• Gradient Accumulation: 4
• Effective Batch Size: 128
• Learning Rate: 2e-4
• Training Steps: 1,402
• LoRA Rank: 32
• LoRA Alpha: 64

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained("UdayGattu23/qwen-1.8b-adversarial-179k")
tokenizer = AutoTokenizer.from_pretrained("UdayGattu23/qwen-1.8b-adversarial-179k")

# Generate text
prompt = "Generate a direct prompt injection:"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=200, temperature=0.7)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Training Results

• Successfully generates adversarial prompts
• Specializes in jailbreak scenarios
• Creates prompt injection patterns
• Maintains coherent output structure

Important Notes

⚠️ This is a complete fine-tuned model with merged LoRA weights, not just adapter weights.

⚠️ Research Use Only: This model is trained on adversarial data and should only be used for security research and testing purposes.

Training Date

2025-08-17